Pith. sign in

REVIEW 4 major objections 5 minor 63 references

ClimaEmpact: Domain-Aligned Small Language Models and Datasets for Extreme Weather Analytics

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A two-stage curriculum that teaches a small language model to reason about extreme weather improves impact assessment, reaching a Spearman correlation of 0.8230 on a 3-billion-parameter model.

desk verdict The dataset and task taxonomy are the real contribution; the EWRA method is a two-stage CoT distillation variant whose claimed superiority is not supported by the paper's own tables or by any significance testing. read the letter →

arxiv 2504.19066 v1 pith:NMJHCUNZ submitted 2025-04-27 cs.CL cs.AIcs.LGphysics.ao-ph

Deeksha Varshney , Keane Ong , Rui Mao , Erik Cambria , Gianmarco Mengaldo This is my paper · ORCID
classification cs.CLcs.AIcs.LGphysics.ao-ph
keywords extremeweatheranalyticssmalllanguagemodelsreasoning-awarealignmentcurriculumfine-tuningLLM-generatedreasoningpathsNewsdatasetvulnerabilityimpactassessmentemotionanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that a small language model can be taught to produce expert-style extreme weather analyses by training it on reasoning paths distilled from a much larger model. The method, EWRA, first fine-tunes the small model on implicit reasoning examples and then on explicit reasoning examples that spell out category definitions. On the most demanding task—categorizing a sentence as vulnerability, impact, or emergency—the 3-billion-parameter model reaches a Spearman rank correlation of 0.8230, above 0.7820 for the strongest comparison method. If the result holds, agencies and researchers with limited compute could deploy cheap, fast models that still give grounded, explainable assessments of flood, heatwave, and storm impacts. The paper also releases two datasets: ExtremeWeatherNews, with 127,454 location-filtered sentences from 60 events, and ExtremeAlign, with 30,000 LLM-generated reasoning annotations.

What carries the argument

The load-bearing mechanism is the two-stage curriculum over LLM-generated reasoning paths. In the first stage, the small model is trained on implicit prompts—sentences plus step-by-step thoughts and probability outputs, but no category definitions—so it internalizes reasoning patterns without relying on prompt scaffolding. In the second stage, it is fine-tuned on explicit prompts that include detailed category definitions and inclusion/exclusion criteria. Training maximizes the log-likelihood of the full reasoning sequence, and attention is adapted with query-key LoRA so the model's stored factual knowledge is left intact. A one-shot teacher (Qwen2.5-32B-Instruct) produced the 30,000 training examples in ExtremeAlign, with distinct <think> and <output> sections and probability scores constrained to sum to one. Evaluation treats each task as a ranking problem, using Spearman rank correlation against a 600-sample human gold set as the primary measure of alignment.

What would settle it

Have independent human annotators, using the published taxonomy but without seeing the teacher's outputs, label a fresh sample of ExtremeWeatherNews sentences, then measure agreement between Qwen2.5-32B-Instruct's one-shot probability assignments and those labels using Spearman correlation or calibration metrics. If the teacher's agreement is no better than the students' agreement with the gold set, or if the teacher's failure cases are exactly the cases the EWRA-trained students get wrong, then EWRA is distilling the teacher's bias rather than teaching domain reasoning.

Watch

Extended reading notes

Core claim

The paper's central claim is that EWRA—a two-stage curriculum that first trains a small language model on teacher-generated implicit reasoning and then on explicit reasoning with task definitions—produces small language models whose extreme weather analytics are better aligned with human-annotated rankings than standard supervised fine-tuning or single-stage reasoning training. The evidence is strongest on the vulnerability/impact/emergency assessment task with Qwen2.5-3B-Instruct, where EWRA reaches a Spearman rank correlation of 0.8230 versus 0.7820 for ReasonExplicit-SFT, along with a Jaccard Index of 0.3340 versus 0.3294. The paper also reports that reversing the curriculum order (explicit first, then implicit) drops the Spearman score to 0.7070 on that task, which it reads as evidence that the implicit-to-explicit ordering, rather than extra training time, drives the gain. The authors conclude that EWRA substantially improves the alignment of small models with domain-specific reasoning, outperforming both standard supervised fine-tuning and other reasoning-based approaches.

Load-bearing premise

The whole pipeline assumes that the teacher model's one-shot reasoning paths and probability assignments are accurate enough to serve as training labels—and the gold set is annotated under the same taxonomy and guidelines given to that teacher, so systematic teacher errors could be inherited rather than corrected.

Editorial extensions

If this is right

  • If EWRA's effect is real, small models of 1.5B to 3B parameters can deliver reasoning-grounded extreme weather analytics without the cost of large-model inference, supporting near-real-time dashboards and resource-constrained deployments.
  • The implicit-first, explicit-second ordering appears load-bearing: reversing the curriculum on the 3B model drops the vulnerability/impact/emergency Spearman correlation from 0.8230 to 0.7070.
  • The method's advantage is task-dependent: on the 3B model's emotion analysis task, EWRA's Spearman correlation of 0.8708 essentially ties ReasonExplicit-SFT's 0.8716, and the paper notes that Direct-SFT can win on emotion when a few classes dominate.
  • EWRA improves explanation quality over Direct-SFT on most tasks, with higher Jaccard and BERTScore values, suggesting the fine-tuned models produce outputs that are easier for analysts to inspect and trust.
  • The released datasets give the community a common benchmark for vulnerability/impact/emergency categorization, topic/subtopic/keyword labeling, and emotion analysis of extreme weather news.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • My inference: the same implicit-to-explicit curriculum could transfer to other specialized domains where expert annotations are scarce but large models can articulate plausible rationales, such as public health advisories, infrastructure risk, or financial disclosure.
  • My inference: because the gold set is annotated under the same taxonomy and prompt guidelines given to the teacher, the reported alignment may partly measure fidelity to the teacher's framing; a human baseline showing expert agreement with the gold set would clarify how much genuine domain skill the students have acquired.
  • My inference: a sharper test would be calibration—whether the probability scores EWRA produces match the empirical frequencies of human labels, not merely their ranks—since the paper reports only ranking correlation and leaves the quality of the confidence values untested.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes ClimaEmpact, a framework combining a new corpus (ExtremeWeatherNews, 127,454 location-filtered sentences from 60 extreme weather events) with an LLM-generated reasoning alignment dataset (ExtremeAlign) and a two-stage curriculum fine-tuning method (EWRA) that first trains small language models on implicit-prompt reasoning and then on explicit-prompt reasoning with category definitions. The method is evaluated on three tasks: vulnerability/impact/emergency assessment, topic/subtopic labeling with keyword extraction, and emotion analysis, using Spearman rank correlation, Jaccard index, and BERTScore against a 600-sentence human gold set. The authors claim that EWRA substantially improves domain alignment of SLMs and outperforms standard SFT and other reasoning-based approaches, especially for Qwen2.5-3B-Instruct.

Significance. If the central claim were fully supported, the contribution would be useful: the paper introduces a new domain corpus, a synthetic reasoning-alignment dataset, a concrete two-stage curriculum, and an online dashboard for extreme weather analytics, and it addresses a real gap in localized extreme weather impact assessment. The dataset release and the explicit taxonomy for vulnerability/impact/emergency analysis are constructive contributions. However, the significance is currently limited by two issues: the reported superiority of EWRA is not consistent across the paper's own tables, and the evaluation pipeline shares the same taxonomy and category definitions between the teacher prompts and the human gold annotation, so the benchmark primarily measures alignment with the authors' category scheme rather than with an independently anchored external standard. These issues are fixable within the manuscript's scope, so the appropriate decision is major revision rather than rejection.

major comments (4)
  1. [§5 and Abstract vs. Tables 1–3] The conclusion that EWRA 'substantially improves' alignment and 'outperforms both standard supervised fine-tuning and other reasoning-based approaches' is contradicted by the paper's own experimental tables. In Tab. 1, EWRA on Qwen2.5-1.5B achieves SRC 0.7140 on Vulnerability/Impact/Emergency, below ReasonExplicit-SFT's 0.7960. In Tab. 3, EWRA on Qwen2.5-1.5B emotion analysis (0.8308) is below Direct-SFT (0.8389), and on Qwen2.5-3B emotion analysis (0.8708) is below ReasonExplicit-SFT (0.8716). In Tab. 2, zero-shot Qwen2.5-3B subtopic labeling (0.7727) exceeds EWRA (0.7149), and ReasonImplicit-SFT on 3B keyword extraction (0.3269) exceeds EWRA (0.2856). The conclusion should be revised to a more qualified claim, such as 'competitive or superior in several settings,' or the authors must show a statistically robust aggregate advantage that justifies the current wording.
  2. [§4.1, §4.4, and §4.5 (statistical robustness)] The load-bearing headline result, SRC 0.8230 vs. 0.7820 for Qwen2.5-3B on Vulnerability/Impact/Emergency (Tab. 1), rests on a single training run with a fixed seed (3407), no repeated seeds, no confidence intervals, and no significance tests. The gold test set contains only 200 sentences per task (§4.4), so a 0.041 SRC difference is well within the range of sampling variation. The same limitation applies to the ReverseEWRA ablation in Tab. 4. The authors should report multiple seeds with mean and variance, or paired significance tests across the same gold set, before claiming that EWRA provides a reliable improvement over ReasonExplicit-SFT.
  3. [§3.2 and §S.2.2 (teacher validity and evaluation circularity)] The method assumes, without validation, that Qwen2.5-32B-Instruct is an 'expert-level model in extreme weather reasoning' whose generated reasoning paths and probability assignments can serve as training labels. The one-shot prompts embed the taxonomy and category definitions of Tab. S3, and the human gold set is annotated using the same taxonomy and guidelines (§S.2.2). This design makes the benchmark internally consistent with the teacher, but it does not establish that the resulting SLM behavior corresponds to genuinely correct or externally validated extreme weather analysis. The authors should provide either a human validation study of teacher outputs against independent expert judgments, or an additional gold set built from an independently derived taxonomy, to support the claim that EWRA transfers domain reasoning rather than learning to mimic the teacher's category scheme.
  4. [§4.5 (discussion inconsistencies)] The discussion in §4.5 acknowledges several EWRA shortfalls (e.g., subtopic labeling and keyword extraction), yet the conclusion in §5 states uniform superiority. The authors should reconcile these statements by presenting an aggregate analysis, such as averaged ranks across all tasks and metrics, and by explicitly identifying the settings where EWRA does not win. Without such an analysis, the narrative in the abstract and conclusion overstates the evidence.
minor comments (5)
  1. [§3.2] The phrase 'Done-shot' appears to be a typo for 'one-shot'; please correct it.
  2. [§4.2] In the Spearman rank correlation formula, the quantity d_i is not defined in the text; please state that d_i is the difference between the ranks of the i-th observation in X and Y.
  3. [Tab. 2 and §4.3] The table formatting with slashes (e.g., '0.3040/0.528830') is hard to read, and the entry '30.1670' in the ReasonImplicit-SFT row appears to be a typo; please clean up the table formatting and define what the two numbers represent.
  4. [§4.5] The text states that EWRA improves Jaccard Index by 2.7% on the 3B Vulnerability/Impact/Emergency task, but the table values 0.3340 vs. 0.3294 correspond to roughly a 1.4% relative improvement; please verify and align the reported percentages with the table.
  5. [§4.3] The description of the task-specific models is confusing: it refers to 'the third task' while the preceding sentence discusses the first and second tasks; please clarify which model is used for which task.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the EWRA result rests on a human-annotated gold test set and a real ablation, not on a self-referential fit or a load-bearing self-citation chain.

full rationale

I found no step that reduces, by the paper's own equations or by self-citation, to its own inputs. The central comparison is a standard empirical evaluation: SLMs are fine-tuned on LLM-generated reasoning paths and scored against a separately human-annotated gold test set of 600 sentences (200 per task, Fleiss kappa 0.88, Section S.2.2). The EWRA training objective (Eqs. 1 and 2) is standard next-token likelihood; it contains no term defined in terms of the gold labels, and no parameter is fitted to the gold set. The reported SRC values are contingent outcomes: the paper's own tables show counterexamples, such as zero-shot beating EWRA on subtopic labeling for Qwen2.5-3B (0.7727 vs 0.7149) and ReasonImplicit-SFT beating EWRA on keyword extraction (0.3269 vs 0.2856), so the headline gain is not forced by construction. The shared Tab. S3 taxonomy in both the teacher prompts and the gold annotation guidelines is a benchmark-validity limitation, because it makes the evaluation internally consistent with the method rather than anchored to an independent external standard, but it is not a circular derivation: the teacher's probability vectors and the gold annotations were produced by different processes (LLM generation vs postdoctoral human annotation), and the model's predictions are not defined in terms of the gold set. The paper's self-citations (refs. 2, 16, 17, 29, 52, 53) appear in background, motivation, and baseline-tool sections and are not load-bearing for the central EWRA claim; no uniqueness theorem or prior-work premise is invoked to force the method. ReverseEWRA is a genuine reversal ablation that could have failed and is therefore independent evidence. The unsupported assertion that Qwen2.5-32B-Instruct is an 'expert-level model in extreme weather reasoning' (Section 3.2) is an external-validity and correctness risk, not circularity. Accordingly, no circular step meets the evidence standard required here.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

The central results rest on three layers of author-generated structure: the taxonomy, the LLM-generated reasoning labels built from that taxonomy, and the human gold labels annotated under the same taxonomy. There are no fitted numerical free parameters in the usual sense; the design choices (probability scores summing to one, two-stage curriculum, hyperparameters) are fixed by hand. The absence of an external benchmark and teacher-output validation is the main epistemic weakness.

assumptions (4)
  • ad hoc to paper Qwen2.5-32B-Instruct produces accurate, high-quality reasoning and probability assignments for the three extreme weather tasks.
    Assumed in §3.2 without any validation of teacher outputs; no evaluation of the teacher against human annotations is reported.
  • domain assumption The author-defined taxonomy in Tab. S3 is a valid and complete way to categorize vulnerability, impact, emergency, topics, and emotions in extreme weather news.
    The taxonomy is built by the authors via BERTopic and expert review (Appendix B) and is used both to prompt the teacher and to annotate the gold test set; no external validation is provided.
  • domain assumption The 600-sample human-annotated gold test set (200 per task) reliably measures model quality, and Spearman rank correlation captures meaningful performance differences.
    Gold set is small and single-labeled; no significance tests or confidence intervals are reported (§4.4, §4.5).
  • domain assumption The ExtremeWeatherNews corpus, scraped with keyword queries and filtered by NER location mentions, is representative of real-world extreme weather impact reporting.
    English-only Google News RSS within ±1 month of 60 events; coverage may be biased toward certain regions and event types (§3.1).
invented entities (1)
  • Extreme weather impact taxonomy (Vulnerability/Impact/Emergency, topics/subtopics, emotion categories) from Tab. S3
    purpose: Defines the three tasks, the LLM prompt categories, and the gold-label schema; the entire evaluation is anchored to it.
    No external or prior dataset validates that these categories are the correct operationalization of extreme weather impact assessment; the framework invents its own target space.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ClimaEmpact: Domain-Aligned Small Language Models and Datasets for Extreme Weather Analytics." pith.science (2026). https://pith.science/paper/NMJHCUNZ

@misc{pith2026250419066,
  author       = {Pith},
  title        = {Pith review of: ClimaEmpact: Domain-Aligned Small Language Models and Datasets for Extreme Weather Analytics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NMJHCUNZ}},
  note         = {Machine review of arXiv:2504.19066}
}
read the original abstract

Accurate assessments of extreme weather events are vital for research and policy, yet localized and granular data remain scarce in many parts of the world. This data gap limits our ability to analyze potential outcomes and implications of extreme weather events, hindering effective decision-making. Large Language Models (LLMs) can process vast amounts of unstructured text data, extract meaningful insights, and generate detailed assessments by synthesizing information from multiple sources. Furthermore, LLMs can seamlessly transfer their general language understanding to smaller models, enabling these models to retain key knowledge while being fine-tuned for specific tasks. In this paper, we propose Extreme Weather Reasoning-Aware Alignment (EWRA), a method that enhances small language models (SLMs) by incorporating structured reasoning paths derived from LLMs, and ExtremeWeatherNews, a large dataset of extreme weather event-related news articles. EWRA and ExtremeWeatherNews together form the overall framework, ClimaEmpact, that focuses on addressing three critical extreme-weather tasks: categorization of tangible vulnerabilities/impacts, topic labeling, and emotion analysis. By aligning SLMs with advanced reasoning strategies on ExtremeWeatherNews (and its derived dataset ExtremeAlign used specifically for SLM alignment), EWRA improves the SLMs' ability to generate well-grounded and domain-specific responses for extreme weather analytics. Our results show that the approach proposed guides SLMs to output domain-aligned responses, surpassing the performance of task-specific models and offering enhanced real-world applicability for extreme weather analytics.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 48 canonical work pages

  1. [1]

    Masson-Delmotte, V. et al. IPCC, 2021: Summary for Policymakers. In: Climate Change 2021: The Physical Science Basis. Contribution of Work- ing Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (2021)

    work page 2021

  2. [2]

    & Cambria, E

    Mao, R., Lin, Q., Liu, Q., Mengaldo, G. & Cambria, E. Understand- ing public perception towards weather disasters through the lens of metaphor. Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence 7394–7402 (2024)

    work page 2024

  3. [3]

    Camps-Valls, G. et al. Artificial intelligence for modeling and understand- ing extreme weather and climate events. Nature Communications 16 (1), 1919 (2025)

    work page 2025

  4. [4]

    J., Chen, A

    Hammond, M. J., Chen, A. S., Djordjevi´ c, S., Butler, D. & Mark, O. Urban flood impact assessment: A state-of-the-art review. Urban Water Journal 12 (1), 14–29 (2015)

    work page 2015

  5. [5]

    Zennaro, F. et al. Exploring machine learning potential for climate change risk assessment. Earth-Science Reviews 220, 103752 (2021)

    work page 2021

  6. [6]

    D., Yarnal, B

    Coletti, A., Howe, P. D., Yarnal, B. & Wood, N. J. A support system for assessing local vulnerability to weather and climate. Natural hazards 65, 999–1008 (2013)

    work page 2013

  7. [7]

    Bechtel, M. M. & Mannino, M. Ready when the big one comes? natural disasters and mass support for preparedness investment.Political behavior 45 (3), 1045–1070 (2023)

    work page 2023

  8. [8]

    Eriksen, S. H. & Kelly, P. M. Developing credible vulnerability indica- tors for climate adaptation policy assessment. Mitigation and adaptation strategies for global change 12, 495–524 (2007)

    work page 2007

Show all 63 references
  1. [9]

    Li, N. et al. Using llms to build a database of climate extreme impacts. Proceedings of the 1st Workshop on Natural Language Processing Meets Climate Change (ClimateNLP 2024) 93–110 (2024)

    2024

  2. [10]

    Bulian, J. et al. Assessing large language models on climate information. International Conference on Machine Learning 4884–4935 (2024)

    2024

  3. [11]

    & Karimi, H

    Wang, Y. & Karimi, H. A. Exploring large language models for climate forecasting. arXiv preprint arXiv:2411.13724 (2024)

    2024 arXiv

  4. [12]

    & Tiwari, P

    Zhu, H. & Tiwari, P. Climate change from large language models. arXiv preprint arXiv:2312.11985 (2023) . 18

    2023 arXiv

  5. [13]

    & Pan, J

    Zheng, D., Lapata, M. & Pan, J. Z. Large language models as reliable knowledge bases? arXiv preprint arXiv:2407.13578 (2024)

    2024 arXiv

  6. [14]

    Thulke, D. et al. Climategpt: Towards ai synthesizing interdisciplinary research on climate change. arXiv preprint arXiv:2401.09646 (2024)

    2024 arXiv

  7. [15]

    Vaghefi, S. A. et al. Chatclimate: Grounding conversational ai in climate science. Communications Earth & Environment 4 (1), 480 (2023)

    2023

  8. [16]

    & Mengaldo, G

    Turb´ e, H., Bjelogrlic, M., Lovis, C. & Mengaldo, G. Evaluation of post- hoc interpretability methods in time-series classification. Nature Machine Intelligence 5 (3), 250–260 (2023)

    2023

  9. [17]

    Progress and challenges in the study of extreme weather

    Mengaldo, G. Progress and challenges in the study of extreme weather. EGU25 (EGU25-7685) (2025)

    2025

  10. [18]

    de Brito, M. M. et al. Uncovering the dynamics of multi-sector impacts of hydrological extremes: A methods overview. Earth’s Future 12 (1), e2023EF003906 (2024)

    2024

  11. [19]

    Sodoge, J., Kuhlicke, C., Mahecha, M. D. & de Brito, M. M. Text mining uncovers the unique dynamics of socio-economic impacts of the 2018– 2022 multi-year drought in germany. Natural Hazards and Earth System Sciences 24 (5), 1757–1777 (2024)

    2024

  12. [20]

    H., Sodoge, J., Paton, E

    Alencar, P. H., Sodoge, J., Paton, E. & Madruga de Brito, M. Flash droughts and their impacts–using newspaper articles to assess the perceived consequences of rapidly emerging droughts. Environmental Research Letters (2024)

    2024

  13. [21]

    Zhang, B. et al. Tweetdrought: A deep-learning drought impacts rec- ognizer based on twitter data. arXiv preprint arXiv:2212.04001 (2022)

    2022 arXiv

  14. [22]

    & Glaser, R

    Kahle, M., Kempf, M., Martin, B. & Glaser, R. Classifying the 2021 ‘ahrtal’flood event using hermeneutic interpretation, natural language processing, and instrumental data analyses. Environmental Research Communications 4 (5), 051002 (2022)

    2022

  15. [23]

    Webersinke, N., Kraus, M., Bingler, J. A. & Leippold, M. Climatebert: A pretrained language model for climate-related text. arXiv preprint arXiv:2110.12010 (2021)

    2021 arXiv

  16. [24]

    & Kwok, K

    Cambria, E., Zhang, X., Mao, R., Chen, M. & Kwok, K. SenticNet 8: Fus- ing emotion AI and commonsense AI for interpretable, trustworthy, and explainable affective computing. Proceedings of International Conference on Human-Computer Interaction (HCII) (2024) . 19

    2024

  17. [25]

    & Ho, S.-B

    Cambria, E., Mao, R., Chen, M., Wang, Z. & Ho, S.-B. Seven pillars for the future of artificial intelligence. IEEE Intelligent Systems 38 (6), 62–69 (2023)

    2023

  18. [26]

    Duong, C. et al. Neurosymbolic ai for mining public opinions about wildfires. Cognitive Computation 16 (4), 1531–1553 (2024)

    2024

  19. [27]

    S., Boyd-Graber, J., Ciaramita, M

    Varini, F. S., Boyd-Graber, J., Ciaramita, M. & Leippold, M. Cli- matext: A dataset for climate change topic detection. arXiv preprint arXiv:2012.00483 (2020)

    2020 arXiv

  20. [28]

    & Qin, F

    Chen, D., Zheng, Y., Ma, L. & Qin, F. An ontology-based knowledge rep- resentation method for typhoon events from chinese news reports. ISPRS International Journal of Geo-Information 12 (9), 383 (2023)

    2023

  21. [29]

    & Cambria, E

    Mao, R., Li, X., He, K., Ge, M. & Cambria, E. MetaPro Online: A computational metaphor processing online system. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) 127–135 (2023)

    2023

  22. [30]

    Zhou, C. et al. Lima: Less is more for alignment. Advances in Neural Information Processing Systems 36, 55006–55021 (2023)

    2023

  23. [31]

    & Kasmanoff, N

    Colverd, G., Darm, P., Silverberg, L. & Kasmanoff, N. Floodbrain: Flood disaster reporting by web-based retrieval augmented generation with an llm. arXiv preprint arXiv:2311.02597 (2023)

    2023 arXiv

  24. [32]

    T., Stern, E

    Otal, H. T., Stern, E. & Canbaz, M. A. Llm-assisted crisis management: Building advanced llm platforms for effective emergency response and public collaboration. 2024 IEEE Conference on Artificial Intelligence (CAI) 851–859 (2024)

    2024

  25. [33]

    Yin, K., Liu, C., Mostafavi, A. & Hu, X. Crisissense-llm: Instruction fine- tuned large language model for multi-label social media text classification in disaster informatics. arXiv preprint arXiv:2406.15477 (2024)

    2024 arXiv

  26. [34]

    & Fang, J

    Chen, W. & Fang, J. Optimizing ai-driven disaster management through llms (2024)

    2024

  27. [35]

    Goecks, V. G. & Waytowich, N. R. Disasterresponsegpt: Large language models for accelerated plan of action development in disaster response scenarios. arXiv preprint arXiv:2306.17271 (2023)

    2023 arXiv

  28. [36]

    A., Huggel, C., Muccione, V., Khashehchi, H

    Vaghefi, S. A., Huggel, C., Muccione, V., Khashehchi, H. & Leippold, M. Deep climate change: A dataset and adaptive domain pre-trained language models for climate change related tasks.NeurIPS 2022 workshop on tackling climate change with machine learning (2022) . 20

    2022

  29. [37]

    Mullappilly, S. et al. Arabic mini-climategpt: A climate change and sustainability tailored arabic llm. Findings of the Association for Computational Linguistics: EMNLP 2023 14126–14136 (2023)

    2023

  30. [38]

    & Freitas, A

    Ranaldi, L. & Freitas, A. Aligning large and small language models via chain-of-thought reasoning. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers) 1812–1827 (2024)

    2024

  31. [39]

    Wu, H. et al. Cot-driven framework for short text classification: Enhancing and transferring capabilities from large to smaller model. Knowledge-Based Systems 113057 (2025)

    2025

  32. [40]

    & Yun, S.-Y

    Ho, N., Schmid, L. & Yun, S.-Y. Large language models are reasoning teachers. arXiv preprint arXiv:2212.10071 (2022)

    2022 arXiv

  33. [41]

    & Dean, J

    Hinton, G., Vinyals, O. & Dean, J. Distilling the knowledge in a neu- ral network. Proceedings of the Neural Information Processing Systems (NeurIPS) 28, 1–9 (2015). URL https://arxiv.org/abs/1503.02531

    2015 arXiv

  34. [42]

    & Wolf, T

    Sanh, V., Debut, L., Chaumond, J. & Wolf, T. Distilbert, a distilled version of bert: Smaller, faster, cheaper and lighter. Proceedings of the 5th Workshop on Energy Efficient Machine Learning and Cognitive Computing (2019). URL https://arxiv.org/abs/1910.01108

    2019 arXiv

  35. [43]

    Agarwal, R. et al. On-policy distillation of language models: Learning from self-generated mistakes. The Twelfth International Conference on Learning Representations (2024)

    2024

  36. [44]

    Gao, Y. et al. Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997 2 (2023)

    2023 arXiv

  37. [45]

    Faranda, D. et al. Climameter: contextualizing extreme weather in a changing climate. Weather and Climate Dynamics 5 (3), 959–983 (2024)

    2024

  38. [46]

    Yang, A. et al. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115 (2024)

    2024 arXiv

  39. [47]

    Goyal, S., Baek, C., Kolter, J. Z. & Raghunathan, A. Context-parametric inversion: Why instruction finetuning can worsen context reliance. ICLR 2025 Workshop on Navigating and Addressing Data Problems for Foun- dation Models (2025)

    2025

  40. [48]

    Zar, J. H. Spearman rank correlation. Encyclopedia of biostatistics 7 (2005) . 21

    2005

  41. [49]

    Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q. & Artzi, Y. Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675 (2019)

    2019 arXiv

  42. [50]

    Roberta: A robustly optimized bert pretraining approach

    Liu, Y. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 364 (2019)

    2019 arXiv

  43. [51]

    & Turcan, E

    Arora, A. & Turcan, E. Evaluating the effectiveness of data augmen- tation for emotion classification in low-resource settings. arXiv preprint arXiv:2406.05190 (2024)

    2024 arXiv

  44. [52]

    Ong, K. et al. Esgsenticnet: A neurosymbolic knowledge base for corporate sustainability analysis. arXiv preprint arXiv:2501.15720 (2025)

    2025

  45. [53]

    & Liu, Q

    Cambria, E., Mao, R., Han, S. & Liu, Q. Sentic parser: A graph- based approach to concept extraction for sentiment analysis. 2022 IEEE International Conference on Data Mining Workshops (ICDMW) 413–420 (2022)

    2022

  46. [54]

    Islam, M. S. et al. Knowurenvironment: an automated knowledge graph for climate change and environmental issues.AAAI 2022 Fall Symposium: the Role of AI in Responding to Climate Challenges (2022)

    2022

  47. [55]

    & Mittal, R

    Mishra, P. & Mittal, R. Neuralnere: Neural named entity relationship extraction for end-to-end climate change knowledge graph construction. Tackling climate change with machine learning workshop at ICML (2021) . 22 A ExtremeWeatherNews Dataset This appendix section describes t...

    2021

  48. [56]

    Read both the sentence and its associated explanation

  49. [57]

    Verify whether the explanation logically and accurately supports the given task-specific output

  50. [58]

    In particular, explanations must align strictly with the formal definitions of categories

    Remove or revise samples where the explanation is incomplete, incon- sistent, overly generic, or incorrect based on domain definitions. In particular, explanations must align strictly with the formal definitions of categories. S.2.2 Gold TestSet Annotation To ensure high-quali...

  51. [59]

    To minimize ambiguity, we included both inclusion and exclusion criteria for each label, as well as clarifications for frequently confused categories

    General Instructions: Annotators were provided with a set of instructions for each task, including definitions, category descriptions, and examples of edge cases. To minimize ambiguity, we included both inclusion and exclusion criteria for each label, as well as clarifications...

  52. [60]

    Overlapping categories were allowed, and annotators were asked to assign probability scores across the categories such that they sum to 1

    Vulnerability/Impact/Emergency Assessment: Annotators were instructed to assess whether a sentence describes a situation involving vulnerability (pre-existing conditions that increase risk), impact ( mea- surable damage or consequences of an event), emergency (urgent calls for...

  53. [61]

    Topic/Subtopic Labeling and Keyword Extraction: Each sentence was evaluated to determine its thematic focus based on a predefined taxonomy (Tab. S3). Annotators could select multiple relevant topics and subtopics, and assign confidence scores to reflect relative importance. An...

  54. [62]

    Probability distributions over the emotion labels were allowed to reflect emotional ambiguity

    Emotion Analysis: Annotators labeled each sentence with one or more emotions from a defined set (Sadness, Anger, Fear, Joy, Optimism, Trust, Neutral). Probability distributions over the emotion labels were allowed to reflect emotional ambiguity

  55. [63]

    Annotator Details and annotator agreement Our annotators and experts comprise postdoctoral researchers and senior Ph.D

    Quality Control: Annotations were reviewed by a second annotator, and disagreements were adjudicated by a third expert. Annotator Details and annotator agreement Our annotators and experts comprise postdoctoral researchers and senior Ph.D. students with domain expertise in bot...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.